/Tech11h ago

Bill Gurley and Steven Sinofsky argue AI will shift to simplified, edge-focused architectures as cost pressures mirror the dot-com era

They say operational margins will dictate tech stack choices.

1710233417.7K

#28

Original post

Bill Gurley@bgurley#28inTech

Maybe we are moving to the "optimization" phase of the AI game. Lots of things change in this phase. In the dot-com boom, all website were built on Oracle+Sun. Five years later all MySQL + Linux. Eventually the margins matter.

Bill Gurley@bgurley

In addition to enterprises pushing to better understand token costs, I think the consumer models (I use many) are "trying less hard" recently. Meaning, trying to do answer without web search. Angling to give up sooner. I have to believe this is a result of cost optimization.

7:11 PM · Jun 6, 2026 · 17.5K Views

/Tech11h ago

Bill Gurley and Steven Sinofsky argue AI will shift to simplified, edge-focused architectures as cost pressures mirror the dot-com era

They say operational margins will dictate tech stack choices.

1710233417.7K

#28

Original post

Bill Gurley@bgurley#28inTech

Bill Gurley@bgurley

7:11 PM · Jun 6, 2026 · 17.5K Views

Sentiment

Users agree with Gurley's prediction of an AI industry shift toward cost optimization, having seen similar changes in models like Claude and building tools accordingly, though some worry this emphasis reduces quality.

Pos

75.0%

Neg

25.0%

9 comments with sentiment.

Cluster Engagement

Posts from X

Most Activity

VIEWS626LIKES5REPLIES2

Tyler Palmer@tyler__palmer

@bgurley So they get lazy over time… just like us

10h6265

BOOKMARKS1

Madhu Guru@realmadhuguru

@bgurley With ~1B MAUs, consumer models are under pressure to optimize inference costs and compute allocation. Your observation tracks.

On phases - from my time on Gemini

7h3941

Crepe Supreme@crepesupreme

The frame holds, though the timeline may be slower this cycle. In dot-com, Oracle had no answer to MySQL. The frontier responded: Anthropic inference margins went 38% to 70%+ while the commodity floor fell 100x. Volume lapped the savings. Thread on the mechanics:

10h4061

Steven Sinofsky@stevesi

Back then it was also a product issue. The mega products had all their complexity where it didn’t matter for web. Not just a margin issue and paying for what you don’t need but a dev and operational complexity you couldn’t use.

I think that will happen with models too. Google mindset drove the architecture the way Oracle and Sun did. The complexity is where Google needs it for their view of massive scale and concentrated capability.

Google would never have focused on edge or on single device as an architectural focus. Everyone has been on that trajectory.

Bill Gurley@bgurley

1h33610

Rune@droletours

@bgurley They are compute constrained, so they get efficient. It's the rational thing to do. Once/if the constraint lifts optimal behavior changes.

9h272

Remote Capital@remote_capital

@bgurley Gemini is noticeably "stupider" of late. Seems like they're hoping people won't notice the drop in quality to save on input costs but it's clear they're not in an equilibrium yet. Still a work in progress on go-to-market strategy, cost optimization, and maximizing user experience

9h461

Bill Gurley@bgurley

@tyler__palmer 🤣🤣🤣

10h5932

Bill Gurley@bgurley

@droletours Good point. Not just costs. Can’t serve everyone.

8h2352

BlackJack@BlackJackPartII

@bgurley Interesting and makes sense. I’ve noticed the same thing recently on Claude. It pushes back on researching topics and is more active in shaping our discussions.

10h1781

Doug Pierce@dmpierce

@bgurley Ok, Bill. PostgreSQL has always been a better choice than MySQL.

10h228

narendra 🎈 🏃🏻‍♂️@narendra

@bgurley As a rounder on the first top 20 media property using MySQL and Linux, this is true! And there were a lot of inflated valuations for companies hawking Oracle and Sun!

10h201

Anil Murty ⟁@anilmurty_ai

@bgurley I’m betting on it. So much so that I built an open source product to help optimize token efficiency http://TokenJam.dev

10h152

thevegasgroup@TheVegasGroup

@bgurley AI slop none the less.

10h96

financeNoCommentLol@financeNoCommen

@bgurley Exactly. It’s going to be open weight models on user owned hardware

8h74

Haim@h17253

@tyler__palmer @bgurley Optimized laziness takes hard work.

10h161

Jim Zellmer@jimzellmer

Except for those who bootstrapped early, who could only afford NT Server + Sybase along with the Kiva Application server (a great product that changed hands a few times). Admittedly a curious recipe. And of course the evolution of those products through msft, netscape, SUN and today Oracle/java is worthy of analysis.

I remember pricing SUN and being shocked. Of course their sales guys drove fine German cars when making calls. :)

Pre linux and mysql was a time. Yet, we survived.

A colleague who had previous experience with Oracle’s pricing tactics refused to entertain, much less consider pl-sql.

History is a blast. It rhymes.

@stevesi

Finally, making platform bets amidst fast changing times: 1995/1996/1997 was a thing. Many firms chose tech that later imploded.

2h43

Remote Capital@remote_capital

@bgurley @OfficialLoganK @joshwoodward new frustration with the current Gemini set-up between flash/lite/pro and normal vs extended thinking. Frequently getting errors using pro-extended thinking and past benefits are gone. Differences in quality between different models for a given query

9h10

Aungnaing@Aungnaing536639

@stevesi @bgurley The societal implications of digital privacy depend heavily on transparency, accountability, and governance in emerging economies.

1h1